Satya Nadella is telling enterprises they are paying for AI twice — once in tokens, and again in the trade secrets they feed into someone else's model. In a blog post published Monday, the Microsoft CEO warned that customers of proprietary model makers like OpenAI and Anthropic are handing over the exact institutional knowledge those labs need to eventually compete with them. It is a striking position from the chief executive of a company that has invested in both.
Nadella's core claim is that AI usage generates a second, hidden invoice. "You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful," he wrote. "The better you want the model to perform, the more of that knowledge you have to feed it."
“You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful. The better you want the model to perform, the more of that knowledge you have to feed it!”— Satya Nadella, Microsoft CEO
The mechanism, in Nadella's telling, is exhaust. Prompts, tool calls, and — most valuably — the corrections users make when the model gets something wrong are all raw material a model provider can learn from. "Every correction is distilled into institutional know-how," Nadella wrote, describing that feedback as "the kind of knowledge a competitor could never buy." The warning echoes concerns previously voiced by investor Jason Calacanis and Palantir CEO Alex Karp, but Nadella is the first sitting hyperscaler CEO to put it in writing.
Key facts
- 01Nadella published the warning in a Monday blog post, telling enterprises they hand proprietary knowledge to model makers with every prompt and correction.
- 02Open source models accounted for 29% of all traffic routed through Vercel's gateway last month, evidence of the shift Nadella is endorsing.
- 03Solo.io CEO Idit Levine says on-prem open source models deliver roughly 90% of proprietary model capability at far lower cost.
- 04In February, Anthropic accused Chinese open source models of distilling Claude via millions of prompts and urged tighter US export controls.
- 05Solo.io powers the Linux Foundation's Agentgateway project and counts T-Mobile, ADP, and SAP as customers.
He also called out an asymmetry in how model providers treat data. Labs argue fair-use rights to train on public web data, then turn around and restrict customers from distilling their outputs. "While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation," Nadella wrote. He is particularly wary of contracts where providers "reserve the right to learn from customer usage and interaction data." In February, Anthropic accused Chinese open source models of doing exactly this — sending millions of prompts to Claude to train cheaper competitors — and urged Washington to tighten export controls.
Nadella's proposed fix is convenient for Microsoft. He wants enterprises to "retain ownership" of their prompts and feedback by building "proprietary learning environments" on the cloud — data that, for most large companies, already sits in Azure. He also urges an "orchestration layer" so customers can swap models rather than lock in to one provider, which is exactly what AI gateway tools now enable.
Nadella never says "open source" explicitly, but the subtext is unmistakable. And large enterprises are already moving. Idit Levine, founder and CEO of Solo.io, which builds networking and security software for AI systems, says her customers are running the math and reaching a clear verdict. "Can I take an open source model and run it on-prem? It will do almost 90% of what the big one's doing. It will cost way less," she said. "They understand that, and they can control it."
The infrastructure supports her read. Solo.io's technology was selected last year to power the Linux Foundation's Agentgateway project, and the company counts T-Mobile, ADP, and SAP as customers. On the developer side, Vercel and OpenRouter have both reported a surge in open model usage — open source models accounted for 29% of all traffic routed through Vercel's gateway last month, a share that would have been unthinkable at the beginning of the enterprise AI wave.
There are caveats. That 90% capability figure is Levine's characterization of what her customers experience, not a benchmark result, and closing the last 10% gap on frontier tasks — long-horizon agents, complex reasoning, multimodal work — still meaningfully favors proprietary systems. Running open models on-prem also shifts operational burden onto internal MLOps teams, which is a real cost that doesn't show up in per-token pricing.
Microsoft's position here is layered. The company sells Azure capacity to OpenAI, hosts and resells OpenAI models, and increasingly offers open weight alternatives through Azure AI Foundry. If enterprises heed Nadella and move workloads to open models on Azure, Microsoft wins the compute regardless of which model providers lose the tokens. That commercial reality doesn't invalidate the argument — it just clarifies why Microsoft is the one making it publicly.
The line that will get quoted back at OpenAI and Anthropic for months is Nadella's closer: "In consuming intelligence, you are creating intelligence. And what you create should belong to you." For the frontier labs whose commercial terms currently claim rights over customer interaction data, that is a direct shot from their largest infrastructure partner.
Nadella has effectively endorsed the enterprise thesis that open source and on-prem models will absorb an outsized share of workload growth from here — a thesis Hugging Face's Clem Delangue has been making for months. The pitch works on two levels: it protects Microsoft's Azure moat by keeping customer data on Microsoft's cloud, and it puts pricing pressure on the frontier model providers Microsoft depends on but does not fully control. Expect every AI procurement conversation inside a Fortune 500 for the next two quarters to open with a version of Nadella's question — and expect proprietary model contracts to quietly loosen their data-use clauses in response.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
